From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-lj1-f195.google.com (mail-lj1-f195.google.com [209.85.208.195]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by dev.tarantool.org (Postfix) with ESMTPS id 84A49441841 for ; Thu, 26 Mar 2020 16:31:18 +0300 (MSK) Received: by mail-lj1-f195.google.com with SMTP id p14so6365360lji.11 for ; Thu, 26 Mar 2020 06:31:18 -0700 (PDT) Date: Thu, 26 Mar 2020 16:31:15 +0300 From: Konstantin Osipov Message-ID: <20200326133115.GA11411@atlas> References: <7982fc7b062b2424689a990de1f76ca2ff0e4f50.1585053743.git.lvasiliev@tarantool.org> <20200324200216.GA18984@atlas> <178dd6a0-cdee-532c-3d0a-af76062d5f6c@tarantool.org> <20200325084205.GG18984@atlas> <09a887d4-7459-683b-8a10-c3a0d27bc8c3@tarantool.org> <20200326121653.GA6796@atlas> <20200326125419.os5yywcjxylpncdc@tarantool.org> <20200326131954.GC8031@atlas> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20200326131954.GC8031@atlas> Subject: Re: [Tarantool-patches] [PATCH 3/6] iproto: Add negotiation phase List-Id: Tarantool development patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Kirill Yukhin , lvasiliev , alexander.turenko@tarantool.org, tarantool-patches@dev.tarantool.org * Konstantin Osipov [20/03/26 16:19]: > A vshard based cluster with 20 router nodes and 20 storage nodes, > 32 core each, will have 20*32*32*16 connections. These are only internal connections maintained by vshard. Replication adds miniscule 32*20*2 connection to that. Any monitoring will be likely establishing a new connection for every monitoring request, 32*40 connections per second. Any quality of service feature, like separating oltp and analytical traffic, could easily require doubling it, since all prioritized work has to use its own transport infrastructure. Any network partitioning event will lead to avalanche like re-establishment of these connections. One can ignore performance concerns, of course, given there has been a foundation of scrupulous accounting for every bit and every CPU cycle, but not very long. > A 20-40 node cluster is not big by any means. > > -- > Konstantin Osipov, Moscow, Russia -- Konstantin Osipov, Moscow, Russia